File formats
Slide images
PNG / JPEG
Standard raster images. Best for local development and region-of-interest captures.
TIFF / OME-TIFF
Standard pathology scan formats. OME-TIFF preserves microscopy metadata including pixel size when embedded.
Whole-slide scanner formats
| Format | Extension | Library |
|---|---|---|
| Aperio SVS | .svs | OpenSlide |
| Hamamatsu NDPI | .ndpi | OpenSlide |
| 3DHISTECH MRXS | .mrxs | OpenSlide |
Current limitation: these formats upload, but tiling them requires the OpenSlide library, which is not installed in the current local or hosted environments, so tile generation fails with an "OpenSlide is not installed" error. The current tiler also reads the full-resolution image into memory, so only small whole-slide files (test regions) would process even with OpenSlide. Pyramid-aware whole-slide support is not currently scheduled. Until it is, export a region as TIFF, OME-TIFF, or PNG from your scanner software or QuPath.
Annotations
GeoJSON
Standard GeoJSON FeatureCollection with polygon or multipolygon geometries. Each feature should include a label or classification property for region naming.
QuPath exports
GeoJSON exports from QuPath follow the same import path. Region properties map to annotation labels in the viewer.
Import from the slide viewer's Annotations panel.
Sample metadata CSV
| Column | Required | Purpose |
|---|---|---|
external_id | Yes | Sample identifier |
diagnosis, tumor_grade, stage, treatment, response_status | No | Clinical metadata |
group_label, timepoint | No | Cohort comparison fields |
Artifact outputs
Pipeline results under data/pathology/artifacts/{run_id}/:
| Artifact | Contents |
|---|---|
tissue_mask.png | Binary tissue foreground |
tissue_regions.json | Region bounding boxes and labels |
cell_overlay.png | Segmentation visualization |
cells.json | Cell centroids, areas, confidence |
density_heatmap.png | Spatial density map |
cell_spatial_metrics.csv | Per-cell spatial measurements |
region_metrics.json | Compartment-level statistics |
phenotypes.parquet, marker_summary.json, composition_by_region.json | Multiplex IF per-cell phenotype calls and region composition |
alignment.json, overlay_points.json, region_expression_summary.parquet (or .csv) | Spatial-alignment coordinate transform, overlay points, and region-level expression summary |
cohort_comparison.json, metric_matrix.csv | Cohort group-comparison statistics |
interpretation.json | AI interpretation narrative with metric citations |
| Classifier model + prediction artifacts | Trained object classifier plus per-cell predicted class/probability |
| Pyramidal OME-TIFF | Exported slide for QuPath/Fiji/OMERO |
Tile pyramids live under data/pathology/tiles/{slide_id}/.
Size guidance
| Image type | Approx. size |
|---|---|
| Region PNG (640×480) | < 1 MB |
| OME-TIFF region scan | 10–100 MB |
| Full WSI (40× H&E) | 1–5 GB |
Large WSI files are processed on the server with tile-first architecture. The frontend never loads full-resolution slides into memory.
Spatial transcriptomics
Visium, Xenium, and MERFISH data formats are handled by the Computational Biology area. Pathology provides tissue morphology overlays and region context — not duplicate spatial omics ingestion.